SFMEB

SFMEB identifies differentially expressed (DE) genes from RNA-seq data using a scaling-free minimum enclosing ball outlier-detection approach to mitigate systematic technical effects and varying sequencing depths.

Key Features:

• Scaling-Free Approach: Eliminates reliance on a single scaling factor for normalization, accommodating multiple scaling factors and varying sequencing depths in RNA-seq data.
• Outlier Detection Framework: Reformulates DE detection as an outlier detection task by constructing a minimum enclosing ball in feature space that contains known non-DE genes and flagging genes outside the ball as DE.
• One-Class Classification: Leverages one-class classification principles by using non-DE genes (e.g., housekeeping or conserved orthologous genes) as the single class, avoiding the need for DE training examples.
• Cross-Species Applicability: Applies to cross-species gene expression comparisons without requiring normalization steps typically needed for comparative RNA-seq analyses.
• Robustness and Performance: Demonstrates robust performance in simulations and real data analyses across heterogeneous settings and biological replicates, with reported superior performance relative to existing methods.

Scientific Applications:

• Biological Research: Identifying DE genes within and between species to inform studies of gene function and regulation.
• Medical Research: Detecting genes with altered expression profiles to support biomarker discovery for diseases.
• Comparative Genomics: Enabling cross-species expression studies by obviating normalization steps in comparative analyses.

Methodology:

Constructs a minimum enclosing ball in feature space containing known non-DE genes and identifies DE genes as those located outside the ball, treating DE detection as an outlier/one-class classification problem.